Introduction

According to the global status report from the World Health Organization (WHO), alcohol consumption is the seventh risk factor for death and causes a 5.1% loss of disability-adjusted life years (DALYs) [1]. Approximately 240 million adults worldwide are estimated to suffer from alcohol use disorder (AUD), including alcohol dependence [2]. Alcohol dependence is the inability to control alcohol use and represents a condition in which a person has a craving for or a physical need to consume alcohol, despite its negative impact on their health and family. Alcohol dependence not only is a mental disorder but also contributes to physical diseases. With a high relapse rate, alcohol dependence has become a main reason for the large disease burden worldwide caused by alcohol use [3]. A systematic review indicated that the current prevalence of alcohol dependence in mainland China for males and females was 4.4% and below 0.2%, respectively [Baseline measurements

Demographic information, including age, education years and smoking status, was collected by a series of self-conducted questionnaires. Clinical characteristics, including age at first drink and number of drinking years, were also measured in the baseline investigation. Alcohol dependence was measured by a Chinese version of the Alcohol Use Disorder Identification Test (AUDIT), which is a short screening tool that made up of 10 questions with total scores ranging from 0 to 40 [29]. Different language versions (including the Chinese version) of AUDIT have been validated [30].

MRI acquisition

A 3.0 T scanner (Achieva; Philips, Amsterdam, the Netherlands) equipped with an Invivo HD 8-channel high-resolution head coil was used to capture MRI images during early abstinence of alcohol. To minimize scanner noise and head movement, we used earplugs and foam padding. Before scanning, participants were told not to fall asleep but to rest quietly with their eyes closed and to empty their minds (checked with participants while finishing the MRI scans).

BOLD signals were obtained with an echo-planar imaging (EPI) sequence at rest for approximately 8 min. Thirty-eight axial slices and 240 volumes covering the entire brain were acquired with repetition time (TR) = 2000 ms, echo time (TE) = 30 ms; flip angle = 90°; matrix = 256 × 256; field of view (FOV) = 24 × 24 cm2; and voxel size = 3.75 × 3.75 × 4 mm3. T1-weighted anatomical images of high resolution with 188 contiguous axial 1 mm thick slices were obtained with TR = 8.1 ms, TE = 3.7 ms; flip angle = 7°; matrix = 256 × 256; and FOV = 25.6 × 25.6 cm2.

Outcomes

After the baseline investigation, alcohol-dependent patients were followed up at 1, 3 and 6 months. During the follow-up interview, all participants were asked the same question: “Have you had alcohol since your last discharge from the detoxification hospitalization?” Participants who replied “yes” were further interviewed for their first redrink time (month) and alcohol use information of the most severe month using a revised Chinese one-month version of AUDIT [31], which allows us to estimate the severity of alcohol use. The timeframe of the revised Chinese one-month version of AUDIT was modified to 1 month. According to previous studies, the optimal cutoff of AUDIT scores to classify alcohol use disorders was eight scores [29, 30]. In follow-up interviews, detoxicated alcohol-dependent patients were included in the relapse group if the AUDIT scores were ≥ 8 and the remaining alcohol-dependent patients were assigned to the nonrelapse group.

Preprocessing of image data

Preprocessing of baseline imaging data was carried out using Data Processing Assistant for Resting-State fMRI (DPARSF) [32] software based on MATLAB (MathWorks, Natick, MA, USA). Image data with head motion parameters ≥1.5 mm or rotation ≥1.5° were excluded. Power et al. calculated head motion (framewise displacement) and found no significant difference between groups [33]. First, the first ten time points were discarded. Second, slice timing, realigning, normalization and resampling were performed in Montreal Neurological Institute (MNI) space at a resolution of 3 × 3 × 3 mm3. Third, we regressed out the nuisance covariates that contained Friston-24 motion parameters, cerebrospinal fluid and white matter [34]. The remaining time series were filtered through a bandpass filter within a frequency range of 0.01 to 0.1 Hz [35] and then smoothed with the Gaussian kernel (full width half maximum: 6 × 6 × 6 mm).

Statistical analyses

ReHo and fALFF analyses

Two-sample t-test were performed on the baseline ReHo maps and fALFF between the alcohol-dependent patients and HCs as well as between the relapse group and the nonrelapse group using DPABI [36]. Age, education years and smoking status were treated as covariates. A corrected significance at the voxel level was set at p <  0.001 for multiple comparisons adjusted by the Family Wise Error (FWE) based on Gaussian Random Field. Clusters were defined in the Automated Anatomical Labeling (AAL) atlas [37].

FC analyses

We defined the brain regions showing significant differences in the ReHo or fALFF between the alcohol-dependent patients and HCs or between the relapse group and the nonrelapse group as seeds to investigate FC indicators that may be associated with alcohol dependence or relapse using two-sample t-test analysis. Smoking status, education years and age were corrected as covariates to control the confounding effect. Statistical inferences were the same as in the ReHo and fALFF analysis.

Logistic regression

To determine the neuroimaging indicators independently associated with alcohol dependence, a binary logistic regression analysis was conducted. The results showed that the FC, ReHo and fALFF values were significantly different between the alcohol dependence group and HCs in the aforementioned two-sample t-test analyses, where age, education years and smoking status were treated as independent variables. The logistic regression analysis to explore the rs-fMRI predictors of alcohol dependence relapse focused on the FC, ReHo and fALFF values that were significantly different between the relapse and nonrelapse groups. In addition, baseline AUDIT scores, first drinking age, years of drinking, average daily consumption in the last month, consumption at the last drinking, age, education years and smoking status were treated as potential confounding variables. The “conditional forward” model was implemented in both binary logistic regression analyses. The area under the receiver operating characteristic curve (AUC) was calculated to show the classifying or predictive power of the regression models.

The 95% confidence intervals (95% CIs), means and standard deviation (SD) were used to describe the demographic variables and outcomes. Student’s t-test and χ2 test were used to calculate the differences between groups for quantitative data and categorical data, respectively. Logistic regression analyses based on a two-tailed α level of 0.05 were conducted in SPSS 26.0 (IBM Corp., Armonk, NY, USA).

Results

Subject characteristics

The age (mean ± SD: alcohol dependence: 39.97 ± 9.00; HC: 38.03 ± 9.53) and education years (alcohol dependence: 12.76 ± 3.52; HC: 13.90 ± 3.58) were not significantly different (age: t = 1.22, p = 0.22; education years: t = − 1.86, p = 0.07) between the alcohol-dependent patients and HCs. In comparison with the HCs (38.24%), the rate of smoking among the alcohol-dependent patients (91.18%) was significantly higher (χ2 = 41.73, p <  0.001). According to the baseline investigation, the mean AUDIT score of the alcohol-dependent patients was 28.13 ± 8.22. At the six-month follow-up investigation, 67 (98.53%) alcohol-dependent patients finished the follow-up research, and 35 (52.24%) of them relapsed. Between the relapse group and the nonrelapse group, differences in smoking status (relapse: 88.57%; nonrelapse: 93.75%; χ2 = 0.10, p = 0.75), age (relapse: 40.97 ± 9.18; nonrelapse: 38.97 ± 8.95; t = 0.90, p = 0.37), education years (relapse: 12.51 ± 3.41; nonrelapse: 13.00 ± 3.72; t = − 0.56, p = 0.58), first drinking age (relapse: 18.71 ± 5.16; nonrelapse: 20.88 ± 5.45; t = − 1.67, p = 0.10), years of drinking (relapse: 22.49 ± 8.94; nonrelapse: 18.84 ± 8.31; t = 1.72, p = 0.09), average daily consumption in the last month (relapse: 171.94 ± 61.31; nonrelapse: 143.39 ± 79.05; t = 1.66, p = 0.10), consumption at the last drinking (relapse: 238.87 ± 678.15; nonrelapse: 151.01 ± 192.46; t = 0.71, p = 0.48) and baseline AUDIT scores (relapse: 29.02 ± 7.82; nonrelapse: 27.09 ± 8.72; t = 1.04, p = 0.3) were not statistically significant. However, the AUDIT score of the most severe month during the follow-up of the relapse group (25.26 ± 10.23) was significantly higher than that of the nonrelapse group (0.44 ± 1.34) (t = 14.23, p <  0.001). The detailed descriptive subject characteristics are presented in Table 1.

Table 1 Sample Demographics and their Clinical Characteristics at baseline survey and follow-up survey
Full size table

rs-fMRI indicators associated with alcohol dependence

ReHo

Compared with the HCs, the alcohol dependence group showed significant reductions in ReHo in the bilateral postcentral gyrus and the bilateral precentral gyrus (Table 2, Fig. 1).

Table 2 Brain regions showing significant differences in ReHo or fALFF and significantly different FCs between healthy controls and alcohol-dependent patients
Full size table
Fig. 1
figure 1

The functional indicators in the brain regions. (Alcohol dependence < HCs, relapse < nonrelapse). Compared to the HCs, the alcohol dependence group had decreased ReHo in the left postcentral (a), left precentral (b), right postcentral (c) and right precentral (d); decreased fALFF in the left postcentral (e), right precentral (f), right postcentral (g) and fusiform (h); decreased FCs of the left precentral (seed) with the right lingual (i), left middle cingulum (j) and left insula (k); decreased FCs of the right precentral (seed) with the right lingual (l), right insula (m) and left superior temporal (n); decreased FC of the left postcentral (seed) with left insula (o); decreased FC of the right postcentral (seed) with the left superior temporal (p); and decreased FC of the right fusiform (seed) with the right middle cingulum (q). The T map was drawn with P <  0.001 at the voxel level and PFWE < 0.05 at the cluster level

Compared to the nonrelapse group, the relapse group had: decreased FC of the left precentral (seed) with the left cerebellum (r, s, t). (P < 0.002 at the voxel level, PFWE < 0.05 at the cluster level). The color bar represents the voxel T value. HCs: healthy controls. ReHo: reginal homogeneity. fALFF: fractional amplitude of low-frequency fluctuations. FCs: functional connectivities.

Full size image

fALFF

As shown in Table 2 and Fig. 1, alcohol-dependent subjects displayed significant reductions in the right fusiform gyrus, the bilateral postcentral gyrus and the right precentral gyrus.

FC

According to the ReHo and fALFF results, we defined the right fusiform, bilateral postcentral and bilateral precentral regions as seeds for the FC analyses. Compared to the HCs, some alcohol-dependent patients had significantly decreased FCs (Table 2 and Fig. 1), including the FCs of the left precentral (as seed) with the right lingual, left middle cingulum and left insula; those of the right precentral (as seed) with the right lingual, the right insula and the left superior temporal gyrus; those of the left postcentral (seed) with the left insula; those of the right postcentral (seed) with the left superior temporal gyrus; and those of the right fusiform (seed) with the right middle cingulum.

rs-fMRI indicators independently associated with alcohol dependence

Three values, the ReHo of the left postcentral gyrus, the fALFF of the right fusiform gyrus and the FC between the right fusiform gyrus and the right middle cingulum, were independently associated with alcohol dependence according to the logistic regression model, which treated the significantly different FC, ReHo and fALFF values between the alcohol dependence group and HCs in the aforementioned two-sample t-test analyses, and age and education years as potential independent variables. When controlling smoking status as a covariate, the difference between the alcohol dependence group and HCs was also significant (Table 3). The AUC of the logistic regression models that entered the three rs-fMRI indicators was 0.841 (95% CI: 0.776, 0.906) (Fig. 2).

Table 3 Variables independently associated with alcohol dependence in the logistic regression models
Full size table
Fig. 2
figure 2

ROC curve of the logistic regression model to differentiate alcohol dependence by functional brain imaging indicators including: the ReHo of the left postcentral, the fALFF of the right fusiform, and the FC of the right fusiform (seed) with the right middle cingulum. 95%CI: 95% confidence interval; ROC: receiver operating characteristic; ReHo: reginal homogeneity; fALFF: fractional amplitude of low-frequency fluctuations; FCs: functional connectivities; AUC: area under the receiver operating characteristic curve

Full size image

rs-fMRI indicators associated with relapse

ReHo and fALFF

We did not find significant differences in ReHo or fALFF between the relapse group and the nonrelapse group.

FC independently predicts the relapse of alcohol dependence

We did not find significant differences in FCs when defining the bilateral postcentral region, the right fusiform region and the right precentral region as seeds. However, when the left precentral region was treated as a seed, a significant reduction in the baseline FC was observed between the left precentral region and the left cerebellum (T = − 4.445, X = − 27, Y = − 66, Z = − 30, PGRF = 0.002) in the relapse group compared with the nonrelapse group (Fig. 1). Only FC was independently associated with the relapse of alcohol dependence according to the logistic regression; FC, baseline AUDIT scores, first drinking age, years of drinking, average daily consumption in the last month, consumption at the last drinking, age, education years and smoking status were treated as potential independent variables. The AUC of the regression model was 0.774 (95% CI: 0.663, 0.886) (Fig. 3).

Fig. 3
figure 3

ROC curve of the logistic regression model to predict the relapse of alcohol dependence by the FC of the left precentral (seed) with the left cerebellum. 95%CI: 95% confidence interval; FC: functional connectivity; ROC: receiver operating characteristic; AUC: area under the receiver operating characteristic curve

Full size image

Discussion

We carried out data-driven research that utilized fALFF and ReHo analyses to identify the seeds of FC analyses in a study on rs-fMRI marks in early alcohol abstinence in alcohol dependence and relapse. Using this study strategy, we found that the reductions in ReHo of the left postcentral gyrus, fALFF of the right fusiform gyrus, and FC between the right fusiform gyrus and the right middle cingulum were independently associated with alcohol dependence. Furthermore, we found that the FC between the left precentral region and the left cerebellum was decreased in the relapse group, which suggested that reduction of the connection of the frontal lobe to the cerebellum might play a role in the pathogenesis of alcohol dependence relapse and might be a potential biomarker to predict relapse.

The postcentral region is associated with the primary sensory center of the brain [38, 39], while the precentral region is known as the primary motor center of the brain [40]. Both the precentral and postcentral regions are involved in the integration of information related to the sensory, motor, attention, and reward circuits [39, 41,42,

Conclusion

This rs-fMRI study found reductions in the functional indicators of several brain regions involved in cognitive, attention, working memory, and emotion and reward regulation among male alcohol-dependent patients with early alcohol abstinence. We documented a novel finding that the decrease in FC between the precentral gyrus and the cerebellum at early alcohol abstinence was longitudinally associated with alcohol dependence relapse during follow-up and may be a potential biomarker to predict alcohol dependence relapse. These findings are meaningful for further research on the neurobiological etiology and biomarkers of relapse for alcohol dependence.